利用模糊近似推論方式構建通勤者決策行為模式

USING FUZZY APPROXIMATE REASONING PROCESSES IN MODELING COMMUTERS’ DECISION BEHAVIOR: A CASE STUDY

傳統通勤決策行為研究常利用機率性模式處理實際決策不確定性的問題，但近來模糊理論應用於解釋事物存在模稜兩可現象已漸被接受，尤其對於主觀認知與判斷所產生之模糊特性利用模糊理論加以解決已是一主要方法，而在董啟崇、趙祖佑 [1]相關研究中亦證明通勤決策過程中之模糊特性；因此本研究以模糊理論結合近似推論(approximate reasoning)方法為基礎，構建模糊化之旅運行為架構，期望嘗試解釋通勤決策過程中之不確定性，並利用適當之模式來描述模糊通勤決策行為。在實證部分，本研究以平日駕駛小客車通勤於台北──淡水之淡江大學教職員及學生為研究範例之實證對象，利用控制實驗(controlled experiment)方式，以真實的通勤者在模擬的通勤環境下，進行連續數日的決策實驗，對其決策行為過程加以觀察與記錄。本研究利用系統化方式分析出發時間與路徑決策行為，並以模糊化旅運行為架構為基礎，構建六個程序性通勤推論決策模型，解釋通勤行為過程；並以實證樣本為基礎，依目前之條件與限制，在上述六個模式中選取三個，針對決策模型進行模式推論結果與實際決策之符合程度進行實證分析，而結果顯示利用近似推論模式之結果對於實際決策行為之符合度平均達七成以上，此結果說明本研究所構建之模式具有相當程度之適用性。

Probability or stochastic choice models have been widely applied to address the uncertainty of traveler's decision behavior. On the other hand, Zadeh's fuzzy set theory introduced to address the phenomenon of ambiguous events rather than random nature may be suitable for exploring human decision behavior such as travel decisions. This study is an attempt to apply fuzzy reasoning method to study travel behavior with a case of auto-driving commuters’ daily departure time and route choices. Within such framework, driver perceptions of uncertain outcome of attributes affecting their travel choices are due to vagueness rather than randomness. A rule-based reasoning process is therefore applied to model the observed behavior rather than the commonly used utility maximization. Furthermore, a systematic hierarchy approach was implemented to describe commuters’ decision process. Observations were established from a controlled experiment in which real commuters were interacting with a simulated traffic context. Six sequential departure time and route decision models were established based on fuzzy inference concepts and those preliminary observations. Each model was then paired with its respective observations. Excluding those with too few observations, three models were validated and showed promising results with matching ratios (between actual decisions and model outputs) higher than 70%.